maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[MLIR] Add function to create BFloat16 array attribute 

This patch adds a new function `mlirDenseElementsAttrBFloat16Get()`,
which accepts the shaped type, the number of BFloat16 values, and a
pointer to an array of BFloat16 values, each of which is a `uint16_t`
value.

Reviewed By: stellaraccident

Differential Revision: https://reviews.llvm.org/D123981
This commit is contained in:
Ashay Rane 2022-04-19 18:53:29 +00:00 committed by Sean Silva
parent 0f8c626723
commit 25c218be36
3 changed files with 23 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
mlir/include/mlir-c/BuiltinAttributes.h
View File

@ -379,6 +379,8 @@ MLIR_CAPI_EXPORTED MlirAttribute mlirDenseElementsAttrFloatGet(
MlirType shapedType, intptr_t numElements, const float *elements);
MLIR_CAPI_EXPORTED MlirAttribute mlirDenseElementsAttrDoubleGet(
MlirType shapedType, intptr_t numElements, const double *elements);
MLIR_CAPI_EXPORTED MlirAttribute mlirDenseElementsAttrBFloat16Get(
MlirType shapedType, intptr_t numElements, const uint16_t *elements);
/// Creates a dense elements attribute with the given shaped type from string
/// elements.

7
mlir/lib/CAPI/IR/BuiltinAttributes.cpp
View File

@ -474,6 +474,13 @@ MlirAttribute mlirDenseElementsAttrDoubleGet(MlirType shapedType,
const double *elements) {
return getDenseAttribute(shapedType, numElements, elements);
}
MlirAttribute mlirDenseElementsAttrBFloat16Get(MlirType shapedType,
intptr_t numElements,
const uint16_t *elements) {
size_t bufferSize = numElements * 2;
const void *buffer = static_cast<const void *>(elements);
return mlirDenseElementsAttrRawBufferGet(shapedType, bufferSize, buffer);
}
MlirAttribute mlirDenseElementsAttrStringGet(MlirType shapedType,
intptr_t numElements,

16
mlir/test/CAPI/ir.c
View File

@ -936,6 +936,7 @@ int printBuiltinAttributes(MlirContext ctx) {
int64_t ints64[] = {0, 1};
float floats[] = {0.0f, 1.0f};
double doubles[] = {0.0, 1.0};
uint16_t bf16s[] = {0x0, 0x3f80};
MlirAttribute encoding = mlirAttributeGetNull();
MlirAttribute boolElements = mlirDenseElementsAttrBoolGet(
mlirRankedTensorTypeGet(2, shape, mlirIntegerTypeGet(ctx, 1), encoding),
@ -974,6 +975,9 @@ int printBuiltinAttributes(MlirContext ctx) {
MlirAttribute doubleElements = mlirDenseElementsAttrDoubleGet(
mlirRankedTensorTypeGet(2, shape, mlirF64TypeGet(ctx), encoding), 2,
doubles);
MlirAttribute bf16Elements = mlirDenseElementsAttrBFloat16Get(
mlirRankedTensorTypeGet(2, shape, mlirBF16TypeGet(ctx), encoding), 2,
bf16s);
if (!mlirAttributeIsADenseElements(boolElements) ||
!mlirAttributeIsADenseElements(uint8Elements) ||
@ -983,7 +987,8 @@ int printBuiltinAttributes(MlirContext ctx) {
!mlirAttributeIsADenseElements(uint64Elements) ||
!mlirAttributeIsADenseElements(int64Elements) ||
!mlirAttributeIsADenseElements(floatElements) ||
!mlirAttributeIsADenseElements(doubleElements))
!mlirAttributeIsADenseElements(doubleElements) ||
!mlirAttributeIsADenseElements(bf16Elements))
return 14;
if (mlirDenseElementsAttrGetBoolValue(boolElements, 1) != 1 ||
@ -1009,6 +1014,7 @@ int printBuiltinAttributes(MlirContext ctx) {
mlirAttributeDump(int64Elements);
mlirAttributeDump(floatElements);
mlirAttributeDump(doubleElements);
mlirAttributeDump(bf16Elements);
// CHECK: dense<{{\[}}[false, true]]> : tensor<1x2xi1>
// CHECK: dense<{{\[}}[0, 1]]> : tensor<1x2xui8>
// CHECK: dense<{{\[}}[0, 1]]> : tensor<1x2xi8>
@ -1018,6 +1024,7 @@ int printBuiltinAttributes(MlirContext ctx) {
// CHECK: dense<{{\[}}[0, 1]]> : tensor<1x2xi64>
// CHECK: dense<{{\[}}[0.000000e+00, 1.000000e+00]]> : tensor<1x2xf32>
// CHECK: dense<{{\[}}[0.000000e+00, 1.000000e+00]]> : tensor<1x2xf64>
// CHECK: dense<{{\[}}[0.000000e+00, 1.000000e+00]]> : tensor<1x2xbf16>
MlirAttribute splatBool = mlirDenseElementsAttrBoolSplatGet(
mlirRankedTensorTypeGet(2, shape, mlirIntegerTypeGet(ctx, 1), encoding),
@ -1094,12 +1101,15 @@ int printBuiltinAttributes(MlirContext ctx) {
float *floatRawData = (float *)mlirDenseElementsAttrGetRawData(floatElements);
double *doubleRawData =
(double *)mlirDenseElementsAttrGetRawData(doubleElements);
uint16_t *bf16RawData =
(uint16_t *)mlirDenseElementsAttrGetRawData(bf16Elements);
if (uint8RawData[0] != 0u || uint8RawData[1] != 1u || int8RawData[0] != 0 ||
int8RawData[1] != 1 || uint32RawData[0] != 0u || uint32RawData[1] != 1u ||
int32RawData[0] != 0 || int32RawData[1] != 1 || uint64RawData[0] != 0u ||
uint64RawData[1] != 1u || int64RawData[0] != 0 || int64RawData[1] != 1 ||
floatRawData[0] != 0.0f || floatRawData[1] != 1.0f ||
doubleRawData[0] != 0.0 || doubleRawData[1] != 1.0)
doubleRawData[0] != 0.0 || doubleRawData[1] != 1.0 ||
bf16RawData[0] != 0 || bf16RawData[1] != 0x3f80)
return 18;
mlirAttributeDump(splatBool);
@ -1123,8 +1133,10 @@ int printBuiltinAttributes(MlirContext ctx) {
mlirAttributeDump(mlirElementsAttrGetValue(floatElements, 2, uints64));
mlirAttributeDump(mlirElementsAttrGetValue(doubleElements, 2, uints64));
mlirAttributeDump(mlirElementsAttrGetValue(bf16Elements, 2, uints64));
// CHECK: 1.000000e+00 : f32
// CHECK: 1.000000e+00 : f64
// CHECK: 1.000000e+00 : bf16
int64_t indices[] = {0, 1};
int64_t one = 1;